The capability of models to faithfully represent the Earth's climate depends on their ability to simulate the full probability density function (PDF), not just the mean or dispersion of climate states. In particular, the extreme ends of the PDF are of great interest, both because extremes tend to have the largest impact on socioeconomic systems and ecosystems and also because extremes are often the most interesting aspect of a distribution. However, extremes are also the most difficult part of the PDF to represent, primarily because they are most sensitive to sampling fluctuations. In this talk, we will provide a quantitative framework for evaluating extreme behavior in precipitation, a non-normal quantity. The methodology will be applied to both observational data and climate model output to determine how it can be used to quantify the fidelity of a given model or to compare the simulations of several models. The advantages and disadvantages of ensembles of model integrations will also be discussed.